Liquid AI Releases Open-Weight d1-3B and d1-omni-600M
Liquid AI has released Open d1, a pair of open-weight multimodal models in its d1 decision model family. d1-3B reads text and images, while d1-omni-600M reads text paired with either an image or audio. Neither model writes text. Instead, each returns calibrated, typed answers in a single forward pass with zero output tokens.
The target is real-time decision-making on the NVIDIA stack: DGX servers, RTX workstations, and Jetson edge boards.
Deployable Today
Both checkpoints are available on Hugging Face, load through Transformers, and have day-one llama.cpp support. The LFM Open License v1.0 allows free commercial use for organizations below $10 million in annual revenue.
Note that d1-omni-600M is an early research release with no published latency figures.
2026 Context: Decision Models Take Shape
The release lands as the industry increasingly separates decision models from conventional generative LLMs. Where most multimodal systems still generate tokens to produce an answer, Liquid AI's d1 family skips generation entirelyβreturning a structured, typed response in one pass. This architectural choice is squarely aimed at latency-sensitive applications in 2026: robotics, on-device agents, industrial automation, and real-time multimodal pipelines where even a few hundred milliseconds of token-by-token decoding is too slow.
With day-one llama.cpp support and a permissive license for smaller organizations, the d1 models are positioned to slot directly into existing NVIDIA-centric inference deployments, from cloud DGX nodes down to Jetson-powered edge devices.
via MarkTechPost